Abstract Mirror tests for ITER have been carried out in JET for over 15 years. During the third JET campaign with the ITER-like wall (2015–2016), comprising a total tokamak plasma exposure duration of 23.4 h and 1027 h of glow discharge cleaning, a new experiment was performed with a specially designed ITER-like test assembly housing six polycrystalline molybdenum mirror samples and featuring trapezoidal entrance apertures simulating the geometry of cut-outs in the diagnostic first wall of the ITER shielding blanket. The assembly was installed on the vacuum vessel wall at the outer midplane, set back radially behind the JET poloidal outer limiters such that the contact with thermal plasma should be largely avoided. The total and diffuse reflectivity of all mirrors was measured in the range 300–2500 nm before and after exposure. Post-exposure studies of mirror surface composition and of surfaces outside and inside the assembly were performed using microscopy, x-ray spectroscopy and ion beam analysis methods. The main results are: (i) no measured degradation of total reflectivity; (ii) diffuse reflectivity increased especially at short wavelengths (below 500 nm) from 1.1 to 2.7% and from 0.8%–1.3% above 1000 nm; (iii) mirrors were coated with a thin co-deposited layer (∼20–30 nm) containing carbon, oxygen and traces of nitrogen, beryllium and metals (Ni, Cr, Fe); (iv) no deuterium was detected; (v) surface composition of the mirror box inner walls was similar to that of the mirrors; (vi) ≲100 nm thick beryllium was the main component on external surfaces of the assembly. These results provide new input to ITER both for the modelling of FM erosion/deposition and for the consideration of requirements for mirror cleaning methods.